Unroll for variables in algorithms

[bjozac]

consider following model

model testDiag
 Real x;
 Integer offset;
 Real arr[3,3];
 Real arr2[3,3]=fill(999,3,3);

 equation
  der(x) = x+time;
  offset = floor(x);

 algorithm
  arr[1,1]:=1.1;arr[1,2]:=1.2;
  arr[2,1]:=2.1;arr[2,2]:=2.2;
  arr[3,1]:=3.1;arr[3,2]:=3.2;
  for k in 1:max(size(arr2)) loop
    arr[k + integer(abs(offset)),k] := arr2[k,1];
  end for;
end testDiag;

model algorithmFor
  testDiag dl;
end algorithmFor;

This will generate the following flat code;

 for k in {1,2,3} loop
    dl.arr[dl.k + integer(Real(abs(dl.offset))), dl.k] := dl.arr2[dl.k, 1];
  end for;

but if your move this part into equation section

  for k in 1:max(size(arr2)) loop
    arr[k + integer(abs(offset)),k] = arr2[k,1];
  end for;

we get

  dl.arr[1 + integer(Real(abs(dl.offset))), 1] = dl.arr2[1,1];
  dl.arr[2 + integer(Real(abs(dl.offset))), 2] = dl.arr2[2,1];
  dl.arr[3 + integer(Real(abs(dl.offset))), 3] = dl.arr2[3,1];

So in the back end we get a compile error due to that dl.k does not exists.

[Adrian Pop]

This is a problem with Prefix.prefixExp, it does lookup of the iterator
and if it finds it in the environment [which of course it does in the
"$for loop scope$"] it will prefix it!

The iterators should never be prefixed, or should also be prefixed in
the for prefix.iterator in .... .

[Adrian Pop]

We talked a bit about this at the meeting yesterday.

For the for iterators in algorithms we should have 2
constantness one for the actual iterator which should
"always" be SCode.VAR and one for the range which
can be SCode.CONST.

If the range is CONST then we can unroll the loop, otherwise
we can't. Example why iterators should be SCode.VAR:

for i in 1:10 loop
 for j in 1:i loop
   ...
 end for;
end for;

In the case above if we set i to have SCode.CONST
then the range 1:i will be constant which it shouldn't be.

I already started looking into this so I will fix it myself.